Asymmetric Fitness of Second-Generation Interspecific Hybrids Between Ciona robusta and Ciona intestinalis

Reproductive isolation is central to speciation, but interspecific crosses between two closely related species can produce viable and fertile hybrids. Two different species of tunicates in the same ascidian genus, Ciona robusta and Ciona intestinalis, can produce hybrids. However, wild sympatric populations display limited gene flow, suggesting the existence of obstacles to interspecific reproduction that remain unknown. Here, we took advantage of a closed culture system to cross C. robusta with C. intestinalis and established F1 and F2 hybrids. We monitored post-embryonic development, survival, and sexual maturation to characterize the genetic basis of simple traits, and further probe the physiological mechanisms underlying reproductive isolation. Partial viability of first and second generation hybrids suggested that both pre- and postzygotic mechanisms contributed to genomic incompatibilities in hybrids. We observed asymmetric fitness, whereby the C. intestinalis maternal lines fared more poorly in our system, pointing to maternal origins of species-specific sensitivity. We discuss the possibility that asymmetrical second generation inviability and infertility emerge from interspecific incompatibilities between the nuclear and mitochondrial genomes, or other maternal effect genes. This work paves the way to quantitative genetic approaches to study the mechanisms underlying genomic incompatibilities and other complex traits in the genome-enabled Ciona model.

separation between the Pacific and Atlantic oceans (Caputi et al. 2007;Bouchemousse et al. 2016b). However, the two species came in contact secondarily, and co-exist in the English Channel, where C. intestinalis is the endemic species, while C. robusta is thought to have invaded the area, in part through human transportation (Zhan et al. 2010;Nydam and Harrison 2011;Sato et al. 2012;Roux et al. 2013;Bouchemousse et al. 2016a). In the contact area, natural hybrids of C. robusta and C. intestinalis were found, but at very low frequencies. Furthermore, the two species displayed limited exchange of alleles (Nydam and Harrison 2011;Sato et al. 2012;Bouchemousse et al. 2016c), suggesting that mechanisms ensuring reproductive isolation largely restrict the expansion of hybrids, as well as gene flow between the two species in the contact region.
Mechanisms ensuring species-specific fertilization are important for prezygotic reproductive isolation (Mayr 1963;Seehausen et al. 2014;Herberg et al. 2018), but successful fertilization between C. robusta and C. intestinalis can routinely be obtained in the laboratory, despite indications that C. intestinalis sperm occasionally fails to fertilize C. robusta eggs (Suzuki et al. 2005;Sato et al. 2014;Bouchemousse et al. 2016a;Malfant et al. 2017). Notably, Ciona adults are self-incompatible hermaphrodites (Harada et al. 2008;Sawada et al. 2020), which spawn their gametes in the open water at dawn. Intrinsic prezygotic isolation would thus involve gamete recognition and/or fertilization success rather than, for example, mating behavior. Nonetheless, prezygotic reproductive isolation in Ciona may not suffice to explain the quasi-absence of natural hybrids and limited gene flow in the wild. Instead, it is thought that postzygotic mechanisms ensure reproductive isolation, including genomic incompatibility in the second generation hybrids. For instance, Sato and colleagues crossed F1 hybrids, produced by forcibly crossing C. robusta and C. intestinalis, and obtained backcrossed BC1 larvae (Sato et al. 2014). However, to our knowledge, the viability and fertility of F2 hybrids, which could provide clues about the physiological origins of the reproductive isolation between Ciona robusta and Ciona intestinalis, has not been reported.
In this study, we took advantage of a simple inland culture system to cross C. robusta and C. intestinalis, and maintain hybrids through multiple generations. We assayed survival, growth and sexual maturation, to further evaluate pre-and postzygotic mechanisms of reproductive isolation between C. robusta and intestinalis. Our observations indicate that F1 and F2 hybrids have reduced fitness compared to C. robusta, with a markedly reduced fertilization success and fitness in specific F2 crosses, suggesting the existence of mechanisms of reproductive isolation. Additionally, we report asymmetric second generation fitness, whereby both hybrids and homotypic animals born from C. intestinalis grandmothers fared poorly. This could be interpreted as inadequacy of our culture system for C. intestinalis, but most likely reflect genomic incompatibilities between the nuclear C. robusta genome and maternally transmitted determinants from C. intestinalis, such as the mitochondrial genotype.

Animals
Wild-type Ciona robusta (C. intestinalis type A) and Ciona intestinalis (C. intestinalis type B) adults were collected in San Diego (CA) and Woods Hole (MA), respectively, and are within the range of known distribution for these species (Nydam and Harrison 2007;Caputi et al. 2007;Bouchemousse et al. 2016b). We confirmed species Figure 1 Crossing animals to make genetic hybrids. (A) Mature animals of Ciona robusta and Ciona intestinalis were collected in San Diego (CA) and Woods Hole (MA), respectively. The animals were in 100 mm Petri dishes. (B) Six animals were dissected to obtain sperm R1 to R6 and eggs R1 to R6 of C. robusta and sperm I1 to I6 and eggs I1 to I6 of C. intestinalis. These sperm and the eggs were homo-and heterotypic crossed to make C. robusta (sperm R x eggs R), RxI hybrid, IxR hybrid and C. intestinalis (IxI) types. identification using established phenotypic criteria (Sato et al. 2012). Sperm and eggs were surgically obtained from mature animals, and used for controlled in vitro fertilizations to produce F1 generation animals, using established protocols (Christiaen et al. 2009). We cultured all animals at 18°(Sanyo, MIR-154), a presumed permissive temperature for both species, as well as for F1 hybrids (Sato et al. 2014;Malfant et al. 2017). Juveniles were kept in Petri dishes at 18°u ntil 28 days post fertilization (dpf). We changed the buffered artificial sea water in the dishes and fed animals every other day. The young animals were transferred into a closed inland culture system at 28 dpf. We measured survival rate by counting the number of live animals in each Petri over time, and measured the size of each living individual from the tip of the siphon to the end of the body. The data were analyzed using Microsoft Office Excel and R. We dissected mature F1 animals to obtain sperm and/or eggs and generated F2 animals by controlled in vitro fertilization.

Algae culture
We essentially followed an established protocol (Joly et al. 2007). We used two strains of microalgae, Chaetoceros gracilis and Isochrysis galbana (aka T.iso) as food for Ciona juveniles and adults, 10 7 to 10 8 cells for each Petri dish, and 10 9 to 10 10 cells for tanks. Stock, starter, and scale-up cultures of algae were kept in 250mL, 500mL and 2L flasks, respectively. Terminal food cultures were kept in 10L carboys. The flasks and carboys were maintained under constant light (Marineland), and were shaken once a day to prevent sedimentation. The cultures were inoculated every 10 to 14 days. Half of the cultures were diluted with autoclaved artificial sea water (Bio-Actif Salt, Tropic Marin) for the next round of cultures. We added 1mL Conway medium (Bouquet et al. 2009;Martí-Solans et al. 2015) and 1g sodium bicarbonate (Sigma Aldrich) per 1L artificial sea water, instead of bubbling CO 2 , to scale up intermediate and terminal food cultures. We added 1mL silicate solution (40g/L metasilicate sodium,

System maintenance
The culture system held 20L glass aquarium tanks (Carolina), 5L polypropylene beakers (Midland Scientific) and 2L polycarbonate aquarium tanks (Eisco), which each could hold 16, 4 and 2 Petri dishes, respectively (Supplemental Figure S1A-C). The 20L tanks, 5L beakers and 2L small tanks were set in an 18°Chamber. Sea water (Bio-Actif Salt, Tropic Marin) was controlled by bio-balls (Biomate, Lifegard Aquatics) seeded with bacteria (BioDigest, Prodibio), and salinity was set at 34 ppt. The 20L tanks were cleaned twice a week, and the 5L beakers and the 2L small tanks were cleaned three times a week. The 20L tanks, 5L beakers and 2L small tanks each efficiently supported animal growth and sexual maturation (Supplemental Figure S1D-E).
Genotyping F1 juveniles were taken from Petri dishes and digested by proteinase (Thermo Fisher Science) to obtain genomic DNA as described previously (Ohta et al. 2010). Oral siphon and sperm were surgically obtained from mature animals, and processed for genomic DNA extraction using the QIAamp DNA Micro Kit (Qiagen), or digested with proteinase. The genomic DNA was used for PCR amplification (35 cycles of 95°30'', 58°30'', 72°1') of target regions with Ex Taq HS DNA polymerase (Takara Bio). The PCR products were purified enzymatically with ExoSAP-IT Express PCR product cleanup (Thermo Fisher Scientific), or by NucleoSpin Gel and PCR clean-up (Macherey-Nagel). PCR products were sequenced by Genewiz. The primers used in this study are summarized in Supplemental Table S1.  with primers designed at Piwi-like, Myl2/5/10, Smad1/ 5/9 and Dync1LI1/2 gene loci from genomic DNAs from sperm of F0 mature animals of C. robusta and C. intestinalis. (B) PCR was done with primers designed at Myl2/5/10 gene locus. Genomic DNA was collected from somatic tissue in oral siphon and sperm of F1 mature adults; C. robusta, RxI hybrid, IxR hybrid and intestinalis. (C) The sequence of PCR products in (B) were read by Myl2/5/10-sequence primer. The asterisks show parts of peaks distinct between C. robusta and C. intestinalis.

Data availability
The authors affirm that all data necessary for confirming the conclusions of the article are present within the article, figures, and tables. Supplemental material available at figshare: https://doi.org/10.25387/ g3.12014826.

RESULTS
Reciprocal crosses between Ciona robusta and C. intestinalis produce hybrids In order to cross Ciona robusta and Ciona intestinalis, we obtained mature animals from San Diego (CA) and Woods Hole (MA), respectively ( Figure 1A). Using six isolated batches of sperm and eggs from each species, we performed homotypic and heterotypic crosses by in vitro fertilization to obtain twenty-four combinations of four types of animals in three separate partial diallels: the parental strains C. robusta and C. intestinalis, and reciprocal F1 hybrids, which we termed RxI, and IxR, for hybrids obtained from C. robusta sperm and C. intestinalis eggs, or C. intestinalis sperm and C. robusta eggs, respectively ( Figure 1B). We obtained hundreds of swimming larvae from each cross (Figure 2A-D), and did not estimate fertilization rates or the proportion of hatched larvae, although this contrasts with previous studies, which suggested that C. robusta oocytes were largely refractory to fertilization by C. intestinalis sperm (Suzuki et al. 2005;Bouchemousse et al. 2016a;Malfant et al. 2018). Further work will be required to determine whether these discrepancies stem from biological and/or experimental differences between studies. We monitored development following hatching, settlement, metamorphosis and initial growth, and did not observe obvious differences between the four types, although this cursory analysis may have missed subtle quantitative variability ( Figure 2E-T). We measured survival rates from 5 to 50 days post fertilization (dpf) by counting the number of animals in each Petri dish ( Figure 3A-B). About 70% of animals in all four conditions survived to 50 dpf ( Figure 3A-B), and an ANOVA did not show significant differences in survival rate between the four types at 26 and 50 dpf, except for between C. robusta and RxI hybrid at 50 dpf ( Figure 3B). Notably, there were no significant differences in the survival rate between F1 RxI and IxR hybrids at 26 or 50 dpf. We monitored the size of animals from 18 dpf to 50 dpf, while keeping the feeding regime constant across conditions Figure 6 Backcrossing to C. robusta eggs. (A) Sperm of (RxR)1 to 6, (RxI)1 to 6, (IxR)1 to 8 and (IxI)1 to 6 were collected from F1 C. robusta, RxI hybrid, IxR hybrid and C. intestinalis mature animals, respectively. Wildtype eggs R7 to R27 were collected from mature animals in CA.     ( Figure 3C-D). Here too, an ANOVA did not reveal significant differences in the size of F1 hybrids at 26 dpf, although size significantly differed between hybrids and C. robusta at 50 dpf ( Figure 3D). Notably, an ANOVA did not show significant size differences between F1 RxI and IxR hybrids at 26 dpf, but showed it at 50dpf. A previous study reported differences in growth rate for hybrid animals of 28 dpf (Malfant et al. 2018). Taken together, these observations suggest that reciprocal first generation hybrids of C. robusta and C. intestinalis are generally as healthy as the parental strains, as they did not display marked differences in post-hatching survival and growth. However, we observed subtle but significant differences between animals obtained from C. robusta or C. intestinalis eggs, whether through homotypic or heterotypic crosses. This observation echoes a previous report that C. robusta eggs confer higher tolerance to challenging water temperatures (Sato et al. 2015), and is a harbinger of asymmetries observed in subsequent crosses (see below). Next, we sought to raise Ciona hybrids to sexual maturity in our experimental facility. In a previous report, Sato and colleagues cultured F1 hybrids in their natural environment, the English Channel, where the two species live in sympatry, and obtained mature animals after a couple of months (Sato et al. 2014). Here, we took advantage of a custom inland culture system to raise and monitor animals through sexual maturation. By 50 dpf, half of the C. robusta individuals were producing sperm, whereas that proportion dropped significantly for the other groups of animals ( Figure 3E). The observation that C. intestinalis F1 animals also fail to produce gametes points to a defect in fitness possibly arising from a specific inadequacy with our culture system, and prevents us from strictly interpreting the lower fitness of RxI and IxR hybrids in terms of genomic incompatibility. We kept these animals until they produced eggs and/or sperm, which we collected surgically, thus sacrificing F1 animals, to test their fertility and obtain F2 animals ( Table 1).

Phenotypes of hybrid adult animals
One obvious difference between parental species is the presence of an orange pigment organ (OPO) at the tip of the sperm duct in C. robusta, but not in C. intestinalis (Millar 1953;Hoshino and Tokioka 1967;Ohta et al. 2010;Sato et al. 2012Sato et al. , 2014Tajima et al. 2019). F1 animals from our parental strains did recapitulate this species-specific trait ( Figure 4A, A', D and D'). For both RxI and IxR hybrids, the majority of animals had OPO at the tip of the sperm duct ( Figure. 4B, B', C, C' and E), in agreement with a previous report (Sato et al. 2014), thus indicating that OPO formation is a dominant trait.
Another character that differs between Ciona species is the color of siphons (Sato et al. 2012), whereby C. intestinalis has yellow and orange pigmentation around the tip of siphons that is lacking in C. robusta ( Figure. 4A, A'', D and D''), although this feature was deemed quite variable and taxonomically unreliable (Brunetti et al. 2015). As for the OPO, the majority of RxI and IxR hybrids displayed a bright red pigmentation at the rim of oral and atrial siphons ( Figure  4B, B'', C, C'' and F), also consistent with a previous report (Sato et al. 2014). The observation that siphon pigmentation displays an overdominant phenotype in hybrids is consistent with its lack of reliability for taxonomic purposes. Further work will be required to determine how proposed species-specific and taxonomically informative traits, such as tubercular prominences in the siphons (Brunetti et al. 2015), which we could not observe clearly, are inherited through generations of hybrids.

Genotyping of hybrid animals
The distribution of variable traits in homo-and heterospecific crosses suggested that RxI and IxR F1 animals are bona fide hybrids. As a complement to phenotypic characterization, and to rule out crosscontaminations during the in vitro fertilization procedure, we sought to perform molecular genetics analyses to assay the distribution of species-specific marker alleles in the different families (Suzuki et al. 2005;Nydam and Harrison 2007). We unsuccessfully tested two primer sets, markers 1 and 2, which were previously used to distinguish C. robusta and C. intestinalis ((Suzuki et al. 2005); Supplemental Figure S2A-C). However, sequence differences between the PCR products distinguished between species-specific alleles (Supplemental Figure S2D). As an alternative, we used a primer set designed at Myosin light chain 2/5/10 (Myl2/5/10; KH.C8.239) locus, which could distinguish C. robusta and C. intestinalis alleles by the size difference of PCR products ( Figure 5A, Supplemental Figure S3A-B). Sequencing amplicons showed conserved 6 th and 7 th exons, but an indel in the 6 th intron that distinguished alleles from different species (Supplemental Figure S3C). We isolated genomic DNA from three F1 juvenile individuals from each type. Six juveniles of either C. robusta or C. intestinalis yielded single bands, albeit of higher molecular weight for the latter (Supplemental Figure S3D). By contrast, six juveniles of either RxI or IxR crosses yielded double bands, showing n■ that these animals had both C. robusta and C. intestinalis Myl2/5/10 alleles, and were indeed hybrids. Consistent with electrophoresis patterns, sequence analysis revealed single alleles for either C. robusta or C. intestinalis, whereas F1 RxI and IxR hybrids produced a mixture of C. robustaand C. intestinalis-specific sequences (Supplemental Figure S3E). Of note, genomic DNA from both somatic tissue and gametes yielded similar results, whereby homotypic C. robusta and C. intestinalis produced single PCR bands in different sizes, while those of both hybrids produced double PCR bands ( Figure 5B). Taken together with the results of phenotypic observations, genotyping data indicated that F1 RxI and IxR animals were bona fide hybrids.
Backcrossing to Ciona robusta eggs Since we could grow F1 hybrids to sexual maturity, we sought to test whether their sperm, which appeared first, could fertilize wildtype C. robusta eggs. For this backcross experiment, we collected sperm from 6, 6, 8 and 6 mature F1 animals of C. robusta, RxI and IxR hybrids, and C. intestinalis, respectively ( Figure 6A, Tables 1-2). On the other hand, we obtained wildtype eggs from 21 (R7-27) mature C. robusta animals. We crossed these sperm and eggs in 78 different combinations (summarized in Table 2). Because F2 (IxI)xR hybrids were potentially equivalent to F1 IxR hybrids, we did not analyze them further. We raised F2 C. robusta animals by crossing sperm from F1 C. robusta (RxR) animals and eggs from C. robusta collected from the wild, and kept F2 C. robusta animals as controls. We counted the proportion of fertilized eggs out of total eggs to score fertilization rates ( Figure 6B-C). The fertilization rates for C. robusta were almost 100%, while the rates dropped and varied between 0-100% in the other crosses ( Figure 6C). Notably, the sperm of F1 RxI hybrid appeared less potent to fertilize C. robusta eggs than that of F1 IxR hybrids, which is reminiscent of previously reported difficulties in using C. robusta eggs in interspecific fertilizations. A heatmap of the fertilization rates showed that there were no infertile eggs from wildtype C. robusta, while sperm from (R1I2)1, (R2I1)3, (I1R2)4, (I1I2)1 and (I1I2)2 might have been sterile ( Figure 6C). While the fertilization rates depend on the quantity and/or quality of sperm, which we did not measure, the range of observed fertilization rates suggested variable compatibilities between specific combinations of sperm and eggs. These observations indicated that F1 hybrids of C. robusta and C. intestinalis can produce fertile sperm capable of fertilizing C. robusta eggs with variable efficacy, which likely constitutes a first, prezygotic, obstacle to interspecific reproduction and gene flow.
Sato and colleagues also successfully obtained mature F1 hybrids, which could be backcrossed to parental species, and the backcrossed BC1 hybrids could develop into seemingly normal larvae (Sato et al. 2014). Likewise, we raised BC1 (RxI)xR and (IxR)xR hybrids at 18°, and allowed them to metamorphose and become young adults by 28 dpf (Figure 6D-E). As a measure of hybrid fitness, we calculated survival rates by counting the number of animals that survived to 28 and 50 dpf relative to the numbers of juveniles at 5 dpf ( Figure 6F). Only half of BC1 (RxI)xR and (IxR)xR hybrid juveniles survived to 28 dpf, compared to almost 90% for C. robusta. Approximately 20% of juveniles of both BC1 hybrids survived to 50 dpf. Both BC1 (RxI)xR and (IxR)xR hybrids had lower survival rates than F2 C. robusta animals, while an ANOVA did not show significant differences in survival rate on 28 and 50 dpf between (RxI)xR and (IxR)xR hybrids. These observations suggest that BC1 hybrid juveniles experience higher mortality rates, consistent with proposed genomic incompatibilities in second generation hybrids (Dobzhansky-Müller Incompatibilities, DMI, (Malfant et al. 2018)). However, in the absence of homotypic C. intestinalis controls, we cannot formally exclude the possibility that the presence of C. intestinalis alleles altered the fitness of hybrid animals in our culturing system, independently of incompatibilities with the C. robusta genome.
As a complement to survival, we measured the body size of BC1 animals at 28 dpf ( Figure 6G). The size of F2 C. robusta juveniles varied between 2 and 4 mm (average = 3.23, SD = 0.75, n = 11), while the size of BC1 (RxI)xR and (IxR)xR hybrids varied between 0.5 and 8 mm ((RxI)xR; average = 3.82, SD = 1.78, n = 47, (IxR)xR; average = 3.70, SD = 2.22, n = 25). This suggested that growth rates are more variable in the BC1 hybrid population, as expected following the segregation of alleles for a likely multigenic trait such as individual growth rate.
n■  Seventeen and ten individuals of (RxI)xR and (IxR)xR hybrids grew to mature adults, respectively, thus allowing us to observe the presence of OPO and the color of their siphons (Figure 7 and Table 3). Except for one individual [(R2I1)2xR8], BC1 (RxI)xR and (IxR)xR hybrids had OPO at the tip of the sperm duct ( Figure 7A and Table 3), which is also consistent with the presence of OPO being a dominant C. robusta trait. Half of the individuals in both BC1 (RxI)xR and (IxR)xR hybrids had red color in the rim of siphons, as did F1 hybrids, while the other half had transparent siphons, the same as normal C. robusta ( Figure 7B). This could be explained considering a single gene, with distinct C. robusta and C. intestinalis alleles, which coexist in F1 hybrids and segregate with a 1:1 ratio in the BC1 hybrid population, because animals heterozygous for C. robusta and C. intestinalis alleles should produce red-colored siphons, as seen in F1 hybrids, while homozygous C. robusta alleles should produce colorless siphons.
Finally, both (RxI)xR and (IxR)xR BC1 hybrids grew and matured to produce sperm and eggs (Table 3 and Supplemental Table S2). The sperm could fertilize C. robusta eggs to produce BC2 hybrids, which survived at least 28 dpf, after which we stopped observations. This indicates that the BC1 hybrids that survive, grow and mature are potentially fertile. This possibility is not incompatible with the existence of DMI. Instead, it is consistent with the existence of defined hotspots of unidirectional introgression observed in wild populations (Roux et al. 2013).

Inbreeding F1 RxI and IxR hybrids
Next, we leveraged the fertility of C. robusta x C. intestinalis offspring to test whether crossing F1 hybrids would yield viable F2 animals, which would in principle provide opportunities for quantitative genetics approaches for the analysis of complex traits. We obtained sperm from 7 and 10 individuals, and eggs from 7 and 11 F1 RxI and IxR mature animals, respectively, and used them for within-type fertilizations ( Figure 8A, Tables 1 and 4). Fertilization rates were significantly higher for IxR hybrids than for RxI hybrids ( Figure 8B). Specifically, crosses between IxR hybrids yielded almost 100% fertilization in 11 trials, except for two combinations, (I6R5)16x(I5R6)18 and (I4R3)17x(I6R5)16, while crosses between RxI hybrids almost invariably failed, except for the (R2I1)7x(R2I1)14 combination ( Figure 8C). The data suggested that the (I6R5)16 F1 adult produced unhealthy gametes, because neither sperm nor eggs yielded productive fertilization. By contrast with backcrossing fertilizations, a limited number of eggs from F1 hybrids produced only hundreds of hatched larvae, thus limiting the numbers of F2 hybrid juveniles in each Petri dish (Table 4). Thus, we calculated metamorphosis rates of F2 hybrids by counting the number of juveniles relative to the number of   7,6,3.5,4,4,2.5,4,4,4,2,1.5,5,8,4,4, 8,6,4,4,3,5,6,4,3,8,2,1,1.5,4,4,4,7,3, 1,6,7,3,7.5,5,1,5,6,1.5,2.5,3,3.5,2,5,3 10,23,27,4,18,15,20,28,14,32,16,18,16, 21,42,33,23,23,9,8,28,30,24,18,13,30, 15,33,21,25,19,25,16,21,8,22,17,9,12 (I3R4) swimming larvae for each fertilization, and could thus evaluate 4 and 10 fertilizations for RxI and IxR crosses, respectively ( Figure  8D). The metamorphosis rates of F2 RxI and IxR hybrids ranged from 0 to 6% and 14%, respectively, which were lower than for C. robusta in regular fertilization (2-80%, average = 25.6%, SD = 15.6, N = 33). Notably, an ANOVA showed significant differences in the metamorphosis rate between F2 RxI and IxR hybrids. Because of low fertilization and metamorphosis rates, we obtained only 7 F2 RxI hybrid juveniles by 5 dpf, compared to 137 F2 IxR hybrid juveniles (Table 4). In total, 3 and 85 juveniles of F2 RxI and IxR hybrids survived to 28 dpf, and displayed normal morphologies, similar to C. robusta ( Figure 8E-F). There were no obvious morphological differences among 28 dpf F2 hybrid individuals between cross types. Survival rates were calculated by counting the number of individuals that survived to 28 dpf and 50 dpf, relative to the number of juveniles at 5 dpf ( Figure  8G). Only 1 juvenile from the (R2I1)9x(R4I3)15 and (R4I3)10x(R5I6)16 crosses survived to 50 dpf, and there were only 2 individuals of F2 RxI hybrids that survived to 50 dpf, compared to 57 F2 IxR hybrid individuals, indicating that F2 hybrids were less viable in the RxI type than in the IxR type. This intriguing observation suggested the existence of asymmetric second generation genomic incompatibilities. However, in the absence of F2 homotypic C. intestinalis controls, we cannot exclude the possibility that the presence of C. intestinalis alleles in the RxI line reduces the fitness of F2 animals in our system, regardless of incompatibilities between genotypes. In either case, the observed asymmetry could involve maternal determinants such as the mitochondrial genome, assuming quasi-exclusive maternal inheritance of the mitochondrial DNA (Nishikata et al. 1987), whereby the C.intestinalis mitochondrial lineage in RxI families lowers their fitness in our culturing system (see discussion). We could measure body sizes for only 3 and 2 F2 RxI hybrid individuals at 28 and 50 dpf, preventing robust statistical analysis. By contrast, 85 and 57 F2 IxR hybrid individuals measured at 28 and 50 dpf showed a range of body sizes similar to that of BC1 hybrids ( Figure 8H). This is also consistent with the notion that body size is a polygenic trait, which displays increased continuous phenotypic variation following alleles segregation of multiple genes in F2. This also suggests that these IxR animals are not obviously subjected to second generation genomic incompatibilities. Finally, although body size is likely multifactorial, i.e., influenced by the environment, especially the availability of food, we surmise that most of the observed variation in controlled laboratory conditions is due to polygenic effects.
One and fifty-three individuals of F2 RxI and IxR hybrids grew to become mature adults, respectively, thus allowing us to observe the presence of OPO and the color of their siphons (Figure 9 and Table 5). The F2 RxI hybrid individual, (R4I3)10x(R5I6), had red color in the rim of siphons. Three-quarters of the individuals in F2 IxR hybrids had OPO at the tip of the sperm duct ( Figure 9E and Table 5), which is also consistent with the presence of OPO being a dominant C. robusta trait. Three-quarters of the F2 IxR hybrids had red color in the rim of siphons, as did F1 hybrids, while the other quarter had transparent siphons, the same as normal C. robusta ( Figure 9F). These proportions are also consistent with mendelian segregation of monogenic traits with dominant species-specific alleles.
Finally, F2 IxR hybrids grew and matured to produce sperm and eggs (Table 5 and Supplemental table S3). The sperm and eggs could fertilize each other to produce F3 IxR hybrids, which survived at least 28 dpf, after which we stopped observations. This indicates that at least the F2 IxR hybrids that survive, grow and mature are fertile, opening future possibilities for inland cultures of hybrid lines.
Following Mendel's laws, the proportions of homo-and heterozygous animals among F2 hybrids should follow a 1:2:1 distribution in the absence of hybrid dysgenesis, inbreeding depression and/or second generation incompatibilities. We analyzed the genotypes at Myl2/5/10 and marker 2 loci for 24 swimming larvae in each of two lines of F2 IxR hybrids, ((I1R2)10x(I2R1)20 and (I3R4)11x(I4R3)21) ( Figure 10). All the PCR amplicons were verified by sequencing, which informed formal genotyping. At the Myl2/5/10 locus, there were 4 and 6 larvae showing homozygous C. robusta alleles, 12 and 13 heterozygous larvae, and 1 and 2 larvae homozygous for the C. intestinalis allele out of 17 and 21 verified samples in (I1R2)10x(I2R1)20 and (I3R4)11x(I4R3)21, respectively ( Figure 10E). The proportion of C. intestinalis genotype was significantly different from the theoretical estimation 25% (P = 1.354e-2 by z-test). By contrast, at the marker 2 locus, there were 2 and 1 larvae homozygous for the C. robusta allele, 13 and 18 heterozygous larvae, and 9 and 5 larvae homozygous for the C. intestinalis allele out of 24 and 24 verified samples in (I1R2) 10x(I2R1)20 and (I3R4)11x(I4R3)21, respectively ( Figure 10F). At this locus, the proportion of C. robusta genotype was significantly different from the theoretical estimation 25% (P = 1.286e-3 by z-test). Biased genotype in Myl2/5/10 showing less C. intestinalis type and marker 2 showing less C. robusta type, suggests that these genes of homozygous type are linked to loci depleted in F2 hybrids populations. Because Myl2/5/10 and marker 2 genes are on different chromosomes and neither are located in the inferred hotspots of introgression (Roux et al. 2013), their allelic distributions might be independent and differentially affected by linkage with incompatible loci. Future work will be required to characterize the genetic underpinnings of genomic incompatibilities between Ciona species, their sensitivity to environmental conditions (including culturing systems), their relationships to documented "hotspots" of introgression (Roux et al. 2013), and their impact on speciation.

DISCUSSION
In this study, we crossed the ascidian species Ciona robusta and Ciona intestinalis to establish hybrid lines, further probe the reproductive isolation of these recently distinguished species, and explore opportunities for quantitative genetics using the genome-enabled Ciona model. Taking advantage of a simple inland culture system, we monitored post-embryonic development and survival, and successfully raised F1 and F2 hybrid and backcrossed animals to maturity. The partial viability of first and second generation hybrids provided insights into the genetics of simple traits, such as the presence of OPO, which appears to be a dominant C. robustaspecific trait. On the other hand, siphon pigmentation showed an n■ overdominant phenotype in hybrids, suggesting more complex genetic interactions, although the trait distribution in F2 hybrids could be explained by allele segregation at one locus. Moreover, simple quantitative traits, such as body size, showed an increased variability in F2 hybrids as expected for polygenic traits following allele segregation. These observations suggest that quantitative genetics approaches could be used to study complex traits that differ between C. intestinalis and C. robusta, such as tolerance to high water temperature (Caputi et al. 2015;Malfant et al. 2017). Despite preliminary evidence of allele segregation in F2 hybrids, the representation of genotype combinations is likely to be biased due to genomic incompatibilities, which might hinder quantitative analysis of polygenic traits. Indeed, our observations suggest that both pre-and postzygotic mechanisms contribute to genomic incompatibilities in hybrids, and thus act as obstacles to interspecific reproduction between these two Ciona species. These observations are consistent with previous reports (Nydam and Harrison 2011;Sato et al. 2012;Bouchemousse et al. 2016a;c). Nonetheless, the incomplete penetrance of first and second generation incompatibilities suggests that certain combinations of C. robusta and C. intestinalis genotypes are viable, which would permit at least low levels of gene flow between populations, and is consistent with the existence of previously reported hotspots of introgression (Roux et al. 2013).
Asymmetric fertilization success in reciprocal interspecific crosses (Turelli and Moyle 2007) was previously observed between Ciona robusta and Ciona intestinalis (Suzuki et al. 2005;Bouchemousse et al. 2016a;Malfant et al. 2018), but the mechanisms remain elusive. On the other hand, asymmetric second generation inviability and infertility points to the mitochondrial genome as the most likely source of reduced fitness in C. intestinalis maternal lineages. Indeed, this asymmetry suggested that mechanisms of genomic incompatibility involve interactions between the nuclear genomes and maternal determinants inherited in a trans-generational manner, such as mitochondrial DNA (Turelli and Moyle 2007;Burton and Barreto 2012;Sloan et al. 2017). However, in the absence of homotypic F2 C. intestinalis animals, we cannot formally rule out the possibility that the C. intestinalis maternal lineage itself causes reduced fitness in our culture system, independently of alleged incompatibilities between the C. robusta and the C. intestinalis genomes. Thoroughly addressing this possibility will require the development of culturing conditions more compatible with C. intestinalis.
Nonetheless, if low C. intestinalis fitness sufficed to explain our results, we would expect to also observe a markedly lower fitness of F1 RxI hybrids compared to F1 IxR hybrids, which emerge from the C. robusta maternal lineage. To the contrary, we could obtain F1 RxI hybrids and raise them to maturity to produce backcrossed animals, which suggested a more pronounced reduction of fitness in second generation RxI hybrids from the C. intestinalis maternal lineage. Moreover, we reasoned that the second generation in the RxI lineage, which necessary follows gametogenesis and possible recombinations between C. robusta and C. intestinalis chromosomes in F1 RxI hybrids, is the first generation where the C. intestinalis mitochondrial genome could encounter homozygous C. robusta alleles in the nuclear genome, which encodes the majority of mitochondrial proteins. For instance, incompatibilities between the nuclear and mitochondrial genomes in hybrids were reported in various organisms, including fungi (Lee et al. 2008;Presgraves 2010), insect (Meiklejohn et al. 2013;Hoekstra et al. 2013), nematode (Chang et al. 2016) and mammals (Ma et al. 2016). For these reasons, and despite the lack of F2 IxI controls, we favor the hypothesis that asymmetric second generation incompatibilities between the C. robusta and C. intestinalis genomes limit the fitness of the RxI hybrid lineage, in a way that depends on the environment, on possible maternal effects genes and/or on specific interactions between the nuclear and mitochondrial genomes. Finally, it is tempting to speculate that these asymmetric incompatibilities provide a mechanistic explanation for the unidirectional introgressions observed in wild populations (Roux et al. 2013).

ACKNOWLEDGMENTS
We are indebted to David Remsen, Josh Rosenthal, Dana Mock-Muñoz de Luna and staff members at MBL, Woods Hole, for extensive support. We thank Régis Lasbleiz for installing the original culture system in our laboratory, and providing detailed instructions for algae and animal culture. We are grateful to Matthew Rockman for critical reading of the manuscript and precious suggestions, insightful discussions and help with the evolutionary theory, concepts and methods of quantitative and population genetics. We thank anonymous reviewers for their help improving the manuscript, in particular for highlighting the potential caveat of lacking C. intestinalis homotypic controls in the F2 generation. This research was supported by NIH/NIGMS award R01 GM096032 and by an MBL Whitman Fellowship to L.C.